Propensity scores · Matching · Causal inference
50 dissertation topics sorted by propensity score method, from nearest-neighbour matching and calipers to IPTW, AIPW, survival matching and sensitivity analysis. Each one comes with a study design, the method to run, a free dataset and a real thesis on something similar.
50 Research questions
10 Sub-fields
13 Free datasets
7 Universities cited
2015 to 2026 Thesis years
By Siddharth Gupta · Updated November 2026
Propensity score matching can help estimate treatment effects from observational data when treated and untreated groups overlap and the analysis accounts for relevant measured confounders. It is used across health, education and economics dissertations.
In my 12 years of running causal models for students, the same three mistakes come up. They match without checking balance, ignore survey weights, or forget that PSM only handles measured confounders. This list matches each topic to the right method from the start.
Every topic below is grouped by method and comes with three things.
A study design that fits the question The method your supervisor will expect A suggested dataset with its access requirements Each topic is also paired with a real thesis from a university repository. Many compare a treated and untreated group in a similar way.
Which matching method do you need? How to use this list Find your situation in the table above and jump to that method. Pick two or three topics in a field you know. Open the dataset and check it measures treatment before the outcome. Read the sample thesis. Look at which covariates it used. Take your shortlist to your supervisor, or to me on a free call. Jump to a sub-field Nearest-neighbour matching Calipers and balance checks Inverse probability weighting Augmented inverse probability weighting (AIPW) Matching with national survey data Matching with survival outcomes Clinical comparisons in hospital data Education programmes Labour and social programmes Sensitivity analysis and alternatives Propensity score research area 01 / 10
Nearest-neighbour matching Start here. Each treated person is matched to the most similar untreated person on the propensity score.
TOPIC 01 Did job training raise earnings, and can matching recover the true effect? Study design Experimental benchmark Statistical method One-to-one nearest-neighbour matching compared with the RCT Suggested datasets LaLonde NSW Job Training Data Open TOPIC 02 Does starting at a community college lower the chance of a degree? Study design Longitudinal cohort Statistical method Nearest-neighbour matching on pre-college traits Suggested datasets NLSY97 Free login TOPIC 03 Do teens who play school sports earn more as adults? Study design Longitudinal cohort Statistical method Nearest-neighbour matching with replacement Suggested datasets Add Health Free login TOPIC 04 Does working during college lower graduation rates? Study design Longitudinal cohort Statistical method Nearest-neighbour matching on family and school factors Suggested datasets NLSY97 Free login TOPIC 05 How much does a college degree add to earnings once people are matched? Study design Longitudinal cohort Statistical method Nearest-neighbour matching on ability and background Suggested datasets NLSY97 Free login PDF details
Propensity score research area 02 / 10
Calipers and balance checks Matching is only as good as its balance. Report standardised mean differences below 0.1 for every covariate.
TOPIC 06 Does extra reading support in kindergarten raise later scores? Study design Longitudinal cohort Statistical method Caliper matching with balance diagnostics Suggested datasets ECLS-K 2011 (NCES) Open TOPIC 07 How much does caliper width change a job training estimate? Study design Experimental benchmark Statistical method Sensitivity of estimates to caliper choice Suggested datasets LaLonde NSW Job Training Data Open TOPIC 08 Do children who get arts instruction score higher in reading? Study design Longitudinal cohort Statistical method Caliper matching with love plots Suggested datasets ECLS-K 2011 (NCES) Open TOPIC 09 Did virtual schooling during COVID-19 hurt child wellbeing? Study design Cross-sectional survey Statistical method Caliper matching with survey weights Suggested datasets IPUMS NHIS Free login TOPIC 10 Do pupils in K-8 schools score higher than those in middle schools? Study design Longitudinal cohort Statistical method Caliper matching on school and family traits Suggested datasets ECLS-K 2011 (NCES) Open PDF details
Propensity score research area 03 / 10
Inverse probability weighting Weighting keeps every person in the sample. It suits large surveys where matching would throw data away.
TOPIC 11 Do adults who lose weight have lower inflammation markers? Study design Cross-sectional survey Statistical method Inverse probability of treatment weighting Suggested datasets NHANES (CDC) Open TOPIC 12 Does supplement use in pregnancy lower anaemia? Study design Cross-sectional survey Statistical method IPTW with stabilised weights Suggested datasets NHANES (CDC) Open TOPIC 13 Did young adults delay cancer screening during COVID-19? Study design Repeated cross-section Statistical method IPTW across survey years Suggested datasets BRFSS (CDC) Open TOPIC 14 Does joining a family support programme lengthen birth spacing? Study design Longitudinal cohort Statistical method IPTW with trimmed weights Suggested datasets NLSY97 Free login TOPIC 15 Are food insecure adults less likely to be screened for colon cancer? Study design Cross-sectional survey Statistical method IPTW with survey design adjustment Suggested datasets NHANES (CDC) Open PDF details
Propensity score research area 04 / 10
Augmented inverse probability weighting (AIPW) AIPW combines a propensity model and an outcome model. The estimate holds if either one is right.
TOPIC 16 Does Medicaid coverage change emergency department use? Study design Panel survey Statistical method Augmented IPW with survey weights Suggested datasets MEPS (AHRQ) Open TOPIC 17 Do adults with chronic gut disease use more care when they gain Medicaid? Study design Panel survey Statistical method Augmented IPW Suggested datasets MEPS (AHRQ) Open TOPIC 18 Does insurance type change the stage at which prostate cancer is found? Study design Registry cohort Statistical method AIPW regression adjustment Suggested datasets SEER Cancer Data (NCI) Free login TOPIC 19 Are adults in states with pharmacist prescribing more likely to get an HIV test? Study design Repeated cross-section Statistical method Augmented IPW Suggested datasets BRFSS (CDC) Open TOPIC 20 Does a housing subsidy reduce household crowding? Study design Cross-sectional survey Statistical method AIPW with survey weights Suggested datasets IPUMS CPS Free login PDF details
Propensity score research area 05 / 10
Matching with national survey data National surveys need weights and design adjustment. Match first, then use the weights in the outcome model.
TOPIC 21 Does heavy screen time in adolescence lower later school results? Study design Longitudinal cohort Statistical method PSM with survey weights Suggested datasets Add Health Free login TOPIC 22 Do Black college students with mental illness get less treatment? Study design Cross-sectional survey Statistical method PSM within survey strata Suggested datasets NSDUH Open TOPIC 23 Does feeling connected to school protect teen mental health? Study design Longitudinal cohort Statistical method PSM with school clustering Suggested datasets Add Health Free login TOPIC 24 Do people who inject drugs in states with syringe laws get tested more? Study design Cross-sectional survey Statistical method PSM with state-level covariates Suggested datasets NSDUH Open TOPIC 25 Do girls in programme villages stay in school longer? Study design Cross-sectional survey Statistical method PSM with cluster-level covariates Suggested datasets DHS Program Data Apply PDF details
Unsure whether your dataset supports matching? Share your proposed treatment, outcome and dataset. We can help assess overlap, pre-treatment covariates and whether matching, IPTW or AIPW suits the question.
Discuss my PSM topic on WhatsApp
Propensity score research area 06 / 10
Matching with survival outcomes Survival topics need matched Kaplan-Meier curves and a Cox model on the matched sample.
TOPIC 26 Did newer treatments improve survival for rare blood cancers? Study design Registry cohort Statistical method Matched Cox model by treatment era Suggested datasets SEER Cancer Data (NCI) Free login TOPIC 27 Did lung cancer survival improve after immunotherapy arrived? Study design Registry cohort Statistical method PSM by diagnosis period with Cox regression Suggested datasets SEER Cancer Data (NCI) Free login TOPIC 28 Are breast cancers found at earlier stages in the tomosynthesis era? Study design Registry cohort Statistical method PSM with stage as outcome and survival follow-up Suggested datasets SEER Cancer Data (NCI) Free login TOPIC 29 Do stroke patients with follow-up care have fewer readmissions? Study design Retrospective cohort Statistical method Matched competing risks model Suggested datasets MIMIC-IV (PhysioNet) Apply TOPIC 30 Does early blood transfusion change survival in trauma patients? Study design Retrospective cohort Statistical method Time-dependent propensity matching Suggested datasets MIMIC-IV (PhysioNet) Apply PDF details
Propensity score research area 07 / 10
Clinical comparisons in hospital data Hospital records are perfect for PSM. Treatments are not random, so matching stands in for a trial.
TOPIC 31 Are older patients on home IV antibiotics readmitted more often? Study design Retrospective cohort Statistical method PSM on age group with outcome regression Suggested datasets MIMIC-IV (PhysioNet) Apply TOPIC 32 Do women and men have different complications after heart procedures? Study design Retrospective cohort Statistical method Sex-matched comparison with PSM Suggested datasets MIMIC-IV (PhysioNet) Apply TOPIC 33 Do patients given early mobility in intensive care leave hospital sooner? Study design Retrospective cohort Statistical method PSM with length of stay outcome Suggested datasets MIMIC-IV (PhysioNet) Apply TOPIC 34 Do older trauma patients do better in specialist units? Study design Retrospective cohort Statistical method PSM on injury severity Suggested datasets MIMIC-IV (PhysioNet) Apply TOPIC 35 Do higher nurse staffing levels lower intensive care deaths? Study design Retrospective cohort Statistical method Generalised propensity score for a continuous dose Suggested datasets MIMIC-IV (PhysioNet) Apply PDF details
Propensity score research area 08 / 10
Education programmes Education is the most common field for PSM dissertations. Students choose programmes, so matching removes selection.
TOPIC 36 Does home computer use raise primary maths scores? Study design Longitudinal cohort Statistical method PSM with growth outcomes Suggested datasets ECLS-K 2011 (NCES) Open TOPIC 37 Does home internet access raise teen school enrolment? Study design Cross-sectional survey Statistical method PSM with survey weights Suggested datasets IPUMS CPS Free login TOPIC 38 Do children taught with more technology score higher in maths? Study design Longitudinal cohort Statistical method PSM on classroom technology use Suggested datasets ECLS-K 2011 (NCES) Open TOPIC 39 Do students who join college clubs persist to graduation? Study design Longitudinal cohort Statistical method PSM on prior achievement Suggested datasets NLSY97 Free login TOPIC 40 Do school-to-work programmes raise early career earnings? Study design Longitudinal cohort Statistical method PSM on high school records Suggested datasets NLSY97 Free login PDF details
Propensity score research area 09 / 10
Labour and social programmes Programme evaluation is where PSM started. These topics test training, schooling and service effects.
TOPIC 41 Do people who lose their job volunteer less? Study design Cross-sectional survey Statistical method PSM with volunteering supplement Suggested datasets IPUMS CPS Free login TOPIC 42 Does employer training raise women’s wages more than men’s? Study design Longitudinal cohort Statistical method PSM by gender subgroup Suggested datasets NLSY97 Free login TOPIC 43 Does secondary schooling lift households out of poverty in South Asia? Study design Cross-sectional survey Statistical method PSM on household wealth Suggested datasets DHS Program Data Apply TOPIC 44 Do children who change schools fall behind? Study design Longitudinal cohort Statistical method PSM on pre-move scores Suggested datasets ECLS-K 2011 (NCES) Open TOPIC 45 Does youth mental health treatment reduce later school problems? Study design Cross-sectional survey Statistical method PSM on symptom severity Suggested datasets NSDUH Open PDF details
Propensity score research area 10 / 10
Sensitivity analysis and alternatives PSM only removes measured bias. Sensitivity analysis shows how strong an unmeasured factor would need to be.
TOPIC 46 Would a hidden factor explain away the link between meth use and pregnancy problems? Study design Cross-sectional survey Statistical method PSM with Rosenbaum bounds Suggested datasets NSDUH Open TOPIC 47 Does good doctor communication lower patient anxiety? Study design Panel survey Statistical method PSM with E-values Suggested datasets MEPS (AHRQ) Open TOPIC 48 Does vaccination lower the risk of long COVID? Study design Cross-sectional survey Statistical method PSM with E-values for unmeasured confounding Suggested datasets IPUMS NHIS Free login TOPIC 49 Do Black mothers with more prenatal visits have better outcomes? Study design Panel survey Statistical method Coarsened exact matching compared with PSM Suggested datasets MEPS (AHRQ) Open TOPIC 50 Is telehealth mental health care as good as in-person care? Study design Panel survey Statistical method Entropy balancing compared with PSM Suggested datasets MEPS (AHRQ) Open PDF details
Free datasets for propensity score projects Open sources are available without an account. Free login requires an account. Apply means access requires approval.
The classic job training dataset used to test matching, with a real experimental benchmark. Use it to practise PSM and compare your estimate with the experimental estimate.
Follows about 9,000 Americans since 1997 through school, college, work-study, training and jobs. Use it for education and labour programme effects.
Follows US teenagers into adulthood with school, activity, media and health data. Use it for youth activity and school climate effects.
Follows US children from kindergarten to grade 5 with services, school type and test scores. Use it for school programme and intervention effects.
US National Health Interview Survey with health, insurance, schooling and long COVID items. Use it for health and child wellbeing effects with survey weights.
US health survey with lab biomarkers, diet, food security and pregnancy data. Use it for health behaviour and biomarker effects.
Yearly survey of 400,000 US adults on prevention, testing and health behaviour by state. Use it for prevention and screening effects.
National US survey of health care use, spending, insurance and patient experience. Use it for insurance, telehealth and care quality effects.
Cancer registry data on stage, treatment, insurance and survival for about half the US population. Use it for cancer treatment and survival comparisons.
Monthly US labour force survey with housing subsidy, internet use and volunteering supplements. Use it for social programme and labour market effects.
Yearly US survey on substance use, mental health and treatment receipt. Use it for mental health service and substance use effects.
Household surveys in 90+ countries with education, wealth and programme exposure. Use it for development programme effects in low-income countries.
Hospital records for over 300,000 patients with treatments, procedures and outcomes. Use it for clinical treatment comparisons in hospital data.
Printable propensity score dissertation guide
Get all 50 topics in one PDF The guide will bring together the questions, study designs, methods and dataset links.
PDF download coming soon How to turn one of these into your own topic Define the treatment and its timing. Covariates must come before treatment.Choose covariates from theory. Include anything that drives both treatment and outcome.Check overlap. Drop people with no comparable match.Report balance. Show standardised mean differences before and after.Run a sensitivity analysis. Report Rosenbaum bounds or an E-value.